Journal of Psychiatric Research 129 (2020) 168–175

Contents lists available at ScienceDirect

Journal of Psychiatric Research

journal homepage: www.elsevier.com/locate/jpsychires

Are there gender differences in prolonged grief trajectories? A registry-sampled cohort study

Marie Lundorff a,b,*, George A. Bonanno c, Maja Johannsen a,b, Maja O’Connor a,b,d a Unit for Bereavement Research, Department of Psychology and Behavioural Sciences, Aarhus University, Aarhus, Denmark b Unit for Psycho-Oncology and Health Psychology, Department of Oncology, Aarhus University Hospital and Department of Psychology and Behavioural Sciences, Aarhus University, Aarhus, Denmark c Department of Clinical Psychology, Columbia University, Teachers College, New York, USA d The Danish National Center for Grief, Copenhagen, Denmark

ARTICLE INFO ABSTRACT

Keywords: Research suggests variation in how grief develops across time, and gender may account for some of this variation. Bereavement However, gender differences in growth patterns of the newly codifiedICD-11 prolonged grief disorder (PGD) are Gender unknown. This study examined gender-specificvariances in grief trajectories in a registry-sampled cohort of 857 Growth mixture modeling spousal bereaved individuals (69.8% female). Participants completed self-report questionnaires of PGD symp­ Prolonged grief toms at 2, 6, and 11 months post-loss. Using Growth Mixture Modeling, four PGD trajectories emerged: resilient Trajectories characterized by low symptoms (64.4%), moderate-stable characterized by moderate symptoms (20.4%), recovery characterized by elevated symptoms showing a decrease over time (8.4%), and prolonged grief characterized by continuous elevated symptoms (6.8%). Similar proportions of men and women comprised the four trajectories. Gender influenced the parameter estimates of the prolonged grief trajectory as men evidenced more baseline symptoms (higher intercept) than women did and a decreasing symptom-level (negative slope), while women showed symptom-increase over time (positive slope). The prolonged grief trajectory captured the largest pro­ portion of probable PGD cases in both genders. Low optimism and low mental health predicted membership in this class. Altogether, the absolute majority of both men and women followed a low-symptom resilient trajectory. While a comparable minority followed a high-symptom prolonged grief trajectory, men and women within this trajectory expressed varying symptom development. Men expressed prolonged grief as an acute, decreasing re­ action, whereas women showed an adjourned, mounting grief reaction. This study suggests that gender may influence symptom development in highly distressed individuals across early bereavement.

1. Introduction for or persistent preoccupation with the deceased accompanied by intense emotional pain, causing significant functional impairment. Bereavement is a universal experience most people undergo during Additionally, the disorder includes a duration criterion, requiring the their life (Holmes and Rahe, 1967), but there are marked individual grief response to persist “for an atypically long period of time following variations in grief severity and duration (Bonanno and Kaltman, 2001). the loss (more than 6 months at a minimum)” (World Health Organi­ Most bereaved individuals manage to adjust and continue normal zation, 2018). functioning after the loss, while a small minority of approximately 10% Given the forthcoming implementation of ICD-11 PGD, knowledge of experiences impairing and pervasive grief reactions (Bonanno et al., variances in risk towards bereavement-related impairment is of partic­ 2002; Prigerson et al., 2009; Shear et al., 2011). This observation ular importance. Gender possibly accounts for some of the observed resulted in the recent codificationof prolonged grief disorder (PGD) as a grief variability. A systematic overview identified female gender as a bereavement-specificmental disorder in the International Classificationof potential risk factor for intense and complicated grief reactions (Burke Diseases (11th version; ICD-11; World Health Organization, 2018). PGD and Neimeyer, 2013), and a meta-analysis demonstrated a small positive describes a pervasive grief response characterized by an intense longing association between female gender and prolonged grief in adults

* Corresponding author. Unit for Bereavement Research, Department of Psychology and Behavioural Sciences, Aarhus University, Aarhus, Denmark. E-mail address: [email protected] (M. Lundorff). https://doi.org/10.1016/j.jpsychires.2020.06.030 Received 11 March 2020; Received in revised form 25 June 2020; Accepted 27 June 2020 Available online 24 July 2020 0022-3956/© 2020 Elsevier Ltd. All rights reserved. M. Lundorff et al. Journal of Psychiatric Research 129 (2020) 168–175 exposed to violent loss (Heeke et al., 2017). Moreover, cross-sectional 2.2. Participants and procedures studies informed by the ICD-11 guidelines showed how female gender was positively associated with PGD symptom-severity (Killikelly et al., Extractions from the Danish Civil Registration System (DCRS) iden­ 2019; Zhou et al., 2020). Meanwhile, in recent meta-analyses, gender tifiedindividuals in the greater metropolitan area of Aarhus, Denmark, did not significantly moderate prevalence rates of disturbed grief who lost a spouse from January 2017 to March 2018. One month post- following neither natural nor unnatural losses (Djelantik et al., 2020; loss, these individuals were contacted and invited to participate in the Lundorff et al., 2017). These seemingly contradictory findings could study. Participants provided written informed consent and received self- reflecthow similar proportions of men and women express problematic report questionnaires at two (T1: M ¼ 2.49 � 0.54), six (T2: M ¼ 6.23 � grief reactions (i.e., no gender difference in grief when studied as a bi­ 0.52), and eleven (T3: M ¼ 11.13 � 0.45) months post-loss. To accom­ nary outcome); however, these reactions might be worse in female modate different needs, participants chose between receiving the diagnostic cases (i.e., a gender difference in grief when studied as a questionnaire by mail or postal service. continuous outcome). As the pathology of ICD-11 PGD symptoms lies in their impairment and continuance (Maciejewski et al., 2016; Maercker 2.3. Measures et al., 2013), longitudinal research seems necessary to explore possible gender differences in symptom-severity, specifically within an ICD-11 Participants completed measures of prolonged grief symptomatology PGD framework. and hypothesized predictors of grief trajectories (i.e., optimism, Growth mixture modeling (GMM) is a robust computational method neuroticism, mental and physical health). Grief symptomatology was to examine changes in psychopathological symptoms over time (Muthen� measured at all time-points (T1-T3), while predictors were measured at and Muth�en, 2018). With GMM, it is possible to differentiate longitu­ baseline (T1). dinal patterns (i.e., trajectories) of PGD from other normative grief re­ actions, such as natural recovery (Bonanno and Malgaroli, 2020), and to 2.3.1. PGD symptomatology investigate systematic variations in these patterns among subgroups, PGD was assessed by mapping the ICD-11 (World Health Organiza­ such as differences between men and women. Earlier GMM studies based tion, 2018) core symptoms (longing, preoccupation) and emotional pain on depression scores have repeatedly demonstrated distinct trajectories symptoms (sadness, guilt, anger, denial, blame, difficultyaccepting the following a loss; a majority of bereaved individuals exhibit minimal death, feeling one has lost a part of one’s self, an inability to experience distress, some express acute distress followed by recovery, while a mi­ positive mood, emotional numbness, difficulty in engaging with social nority show chronic high-level distress (e.g., Bonanno et al., 2005; or other activities) with items from three questionnaires (see Supple­ Galatzer-Levy and Bonanno, 2012; Maccallum et al., 2015). Recent mentary Table 1). The Prolonged Grief-13 (PG-13; Prigerson et al., 2009) studies extend this prior line of research by using grief-specificmeasures and the Inventory of Complicated Grief-Revised (ICG-R; Prigerson and to investigate longitudinal patterns of normative and pathological grief Jacobs, 2001) measure functionally impairing grief symptoms (e.g., “Do (Bonanno and Malgaroli, 2020; Lenferink et al., 2018; Nielsen et al., you feel emotionally numb since your loss?“). Individuals rate how 2018; Smith and Ehlers, 2019; Sveen et al., 2018). disturbing these symptoms are, and higher scores indicate elevated grief. The present study evaluated gender differences in grief trajectories The PTSD Checklist-Civilian Version (PCL-C; Ruggiero et al., 2003) mea­ trough GMM in a large registry-sampled cohort of spousally bereaved sures post-traumatic stress symptoms (PTSS) following potentially individuals. Previous grief-trajectory studies examined symptoms from traumatic events. In the current study, PCL-C items were adjusted to different conceptualizations of pathological grief, such as Persistent refer to “the death” instead of the generic term “stressful experience” (e. Complex Bereavement Disorder (PCBD; American Psychiatric Associa­ g., “Trouble experiencing positive feelings after the death?“). PCL-C was tion, 2013), PGD (World Health Organization, 2018), and Prolonged included to augment the capture of prolonged grief symptoms, and not Grief as per Prigerson et al. (2009). A recent comparative analysis, as a PTSS-specific measure. The mapped PGD items showed excellent however, indicated that trajectories based on the symptoms of PGD from internal consistency (α[T1] ¼ 0.89; α[T2] ¼ 0.90; α[T3] ¼ 0.90). the ICD-11 best-captured change across time (Bonanno and Malgaroli, For the GMM analyses, the two core and the ten emotional pain 2020). Accordingly, the present study relied on these symptoms and symptoms were rated as present/absent and then summed for a total investigated whether men and women differed in a) membership pro­ PGD symptom-severity score (range: 0–12). It was also tested how well portions across trajectories, b) distribution of PGD cases across trajec­ trajectories captured diagnostic PGD cases at T2 and T3. In line with tories, and c) early predictors of trajectory membership. To our earlier research (e.g., Boelen et al., 2019, 2018), probable PGD caseness knowledge, this is the first study to investigate gender differences in required � one core symptom, �one emotional pain symptom, and growth patterns of ICD-11 PGD using GMM. present significant impairment. The dichotomous (yes/no) PG-13 item, “Have you experienced a significantreduction in social, occupational, or 2. Materials and methods other important areas of functioning (e.g., domestic responsibilities)?“, assessed impairment. 2.1. Data 2.3.2. Hypothesized psychosocial predictors Data were collected for The Aarhus Bereavement Study (TABstudy), The Life Orientation Test-Revised (LOT-R; Scheier et al., 1994) a large multi-wave cohort study on natural and prolonged grief reactions measured optimism by rating agreement with ten statements (e.g., “In managed by the last author of this paper [psy.au.dk/grief]. The uncertain times, I usually expect the best”). Higher scores indicate TABstudy follows the General Data Protection Regulation of the Euro­ higher levels of optimism. Neuroticism was captured with its respective pean Union [2016/679], is conducted under the surveillance of the subscale on the NEO Personality Inventory-Revised (NEO-PI-R Short Danish Data Protection Agency [registration number: 2015-57-0002- Version; Costa and McCrae, 1992), consisting of domain-specific per­ 62908-266] and was, due to the involvement of human volunteers, sonality statements (e.g., “I often feel tense and nervous”). Higher scores pre-registered as an observational study at ClinicalTrials.org specify higher levels of neuroticism. The Short-Form Health Survey [NCT03049007]. Data were collected and managed using the research (SF-12; Ware et al., 1996) is a generic measure of health status, electronic data capture (REDCap) tool hosted at Aarhus University addressing different aspects of emotional states and daily activities (e.g., (Harris et al., 2009). “Have you felt downhearted and blue?“, “Have you accomplished less than you would like as a result of your physical health?“). The SF-12 calculates subscores for mental and physical health based on popula­ tion norms with higher scores indicating better health. These covariates

169 M. Lundorff et al. Journal of Psychiatric Research 129 (2020) 168–175 showed good internal validity (α[LOT-R] ¼ 0.78. α[NEO-PI-R] ¼ 0.89. Table 1 α[SF-12] ¼ 0.85). Socio-demographic, loss-related, and psychological characteristics. Total Male Female p sample (N subsample (n subsample (n ¼ 2.4. Statistical analyses ¼ 857) ¼ 259) 598) Gender, n (%) First, the appropriateness of conducting GMM was established Males 259 (30.2) 259 – – through growth curve modeling, where an unconditional single growth Females 598 (69.8) – 598 Age, mean (SD) 70.3 (9.8) 71.7 (9.8) 69.7 (9.8) .006 curve was fitted to the data. Adequate fit (Yu, 2002) was indicated by a � � Educational attainment , n (%) Comparative Fit Index (CFI 0.96), Tucker-Lewis Index (TLI 0.95), Primary 498 (55.8) 136 (52.5) 342 (57.2) .313 root-mean-square error of approximation (RMSEA�0.05), and stan­ education dardized root-mean-square residual (SRMR�0.07). Secondary 337 (39.3) 107 (41.3) 230 (38.5) Second, GMM explored unconditional models with one to five tra­ education Missing 42 (4.9) 16 (6.2) 26 (4.3) jectories (classes) with intercept and slope tested as fixed or random Source of incomeb, n (%) effects. To facilitate convergence in the final model testing, intercept Salary 132 (15.4) 39 (15.0) 93 (15.5) .605 variances were allowed to be random across classes, while slope vari­ Pension 656 (76.5) 197 (76.1) 459 (76.8) ances were fixed.Identification of the optimal number of classes entailed Out of 36 (4.2) 8 (3.1) 28 (4.7) comparison of progressive model solutions using standard fit indices, employment Missing 33 (3.9) 15 (5.8) 18 (3.0) including Akaike Information Criterion (AIC), Bayesian Information Cause of death, n (%) Criterion (BIC), and Sample Size Adjusted BIC (SSA-BIC) for which lower Old age 21 (2.4) 2 (0.8) 19 (3.2) .001 values suggest better model fit. Entropy values range between 0 and 1 Cancer 473 (55.2) 156 (60.2) 317 (53.0) with higher values signifying better classificationaccuracy, and finally, Cardiovascular 99 (11.6) 24 (9.3) 75 (12.5) the Lo-Mendell-Rubin likelihood ratio test (LMR-LRT) and the boot­ disease Dementia 17 (2.0) 11 (4.2) 6 (1.0) strapped LRT (B-LRT) indicate whether adding a class improves model Accident/suicide 23 (2.7) 7 (2.7) 16 (2.7) fit significantly. These indices, explanatory properties, and theoretical Diabetes 8 (0.9) 3 (1.2) 5 (0.8) coherence (Bonanno, 2004; Muthen,� 2003; Nagin and Odgers, 2010) Other 179 (20.9) 41 (15.8) 138 (23.1) informed final model selection. Missing 37 (4.3) 15 (5.8) 22 (3.7) PGD symptom-severityc, M (SD) Third, multiple-group GMM analysis conditioned by the best-fitting T1 3.2 (2.4) 3.1 (2.4) 3.3 (2.4) .212 model with gender as a known-class variable tested the effect of T2 2.6 (2.5) 2.4 (2.6) 2.6 (2.4) .317 gender on trajectories. Wald Chi-Square Tests of parameter equality T3 1.9 (2.2) 1.7 (2.2) 2.0 (2.3) .056 examined the extent to which model parameters (i.e., slope and inter­ Optimism, M (SD) 15.4 (3.9) 15.7 (3.8) 15.2 (4.0) .151 < cept) differed between men and women. Z-tests compared proportions Neuroticism, M 29.9 (8.4) 27.3 (8.0) 31.0 (8.4) .001 (SD) ’ of men versus women in each trajectory. Fisher s Exact Tests evaluated Mental health, M 43.5 (12.1) 45.6 (11.5) 42.5 (12.2) .001 distributional variances of ICD-11 PGD cases across male and female (SD) trajectories. Finally, multinomial regression examined potential pre­ Physical health, M 48.8 (11.1) 49.4 (11.3) 48.6 (11.1) .352 dictors of trajectory membership by nesting theoretically relevant (SD) covariates (optimism, neuroticism, mental health, and physical health) PGD, prolonged grief disorder; SD, standard deviation; T1, time 1 (2 months in the model. Analyses were performed using Mplus, version 8.1 post-loss); T2, time 2 (6 months post-loss); T3, time 3 (11 months post-loss). (Muth�en and Muth�en, 2018), and SPSS, version 26 (IBM Corporation, Statistically significant values indicated in bold font. a 2019). Educational attainment was dichotomized as 1) primary education (primary school, high school, vocational training) and 2) secondary education (college, university). 3. Results b Source of income was categorized as 1) salary for those holding a job, 2) pension for those on early voluntary retirement and those receiving self-financed 3.1. Study population or government-assisted pension, and 3) out of employment (including unem­ ployment benefits,government-sponsored support, social security payments due Consecutive registry-extractions identified a total of 2006 bereaved to sickness). spouses, of whom 857 provided sufficientPGD data for inclusion in the c PGD symptom-severity reflects the summed score of present core and present GMM analysis. Comparisons between those from the total emotional pain symptoms (ranging from 0 to 12). sampling pool not included (N ¼ 1149) and the analyzed sample (N ¼ 857) revealed no significant gender differences (ꭕ2(1) ¼ 0.974, p ¼ 3.2. Growth curve modeling .805), but those not included were older than the analyzed sample (t ¼ 3.839, p < .001). Conventional cut-off (Sullivan and Feinn, 2012) sug­ Growth curve modeling of PGD scores over time with an uncondi­ gest this difference reflects an unsubstantive effect size (d ¼ 0.17; 95% tional growth curve demonstrated excellent data fit(CFI ¼ 0.997; TLI ¼ confidenceinterval [CI] ¼ 0.08-0.26). Number of responses on the PGD 0.991; SRMS ¼ 0.014; RMSEA ¼ 0.065) and found the latent slope measure varied across data-points (T1: N ¼ 839, T2: N ¼ 773, T3: N ¼ variance significant (p < .001). Although the RMSEA was marginally 749), but was neither statistically associated with gender, age, educa­ above the suggested cut-off, the remaining fit indices indicated that tion, source of income, cause of death, nor PGD symptom-severity at GMM was appropriate to explore heterogeneous classes within the data. T1-T3 (ps > .05). The analyzed sample had an average age of 70.3 years (�9.8), 69.8% 3.3. Growth mixture modeling was female, 55.8% had completed primary education, and 76.5% was retired. The male subsample was significantly older, reported better Unconditional models with one to five classes were compared to mental health, and lower neuroticism compared to the female subsam­ identify the best-fittingnumber of classes (Table 2). Information criteria ple (see Table 1), reflecting differences of small effect sizes (ds ¼ (AIC, BIC, SSA-BIC) decreased steadily across consecutive models, 0.21–45; 95% CIs ¼ 0.06-0.61). Men and women also differed regarding indicating an increasingly better model fit. Except for the three-class the cause of death (p ¼ .001), primarily due to a higher proportion of solution, entropy was above 0.8, suggesting high classification accu­ men bereaved by dementia (z ¼ 2.6, p < .01). racy (Clark and Muth�en, 2009). The LMR-LRT varied, but demonstrated

170 M. Lundorff et al. Journal of Psychiatric Research 129 (2020) 168–175

Table 2 3.4.3. PGD case match Fit Indices for one- to five-class models of prolonged grief symptomatic Distribution of PGD caseness at T2 and T3 differed significantly (ps trajectories. < .001) across the identified trajectories (Table 3). At T2, a minority Fit index One-class Two-class Three-class Four-class Five-class (6.8%) of those characterized as resilient met probable PGD caseness,

AIC 9587.890 9447.235 9367.912 9272.219 9220.444 while 84.6% of the male and 67.7% of the female prolonged grief class BIC 9616.411 9490.016 9424.953 9343.521 9306.006 met caseness. Similar results were found at T3, with 45.5% of the male SSA-BIC 9597.356 9461.434 9386.844 9295.885 9248.843 and 74.1% of the female prolonged grief class meeting caseness. The Entropy – 0.818 0.769 0.815 0.827 within-class percentages of caseness decreased from T2 to T3 for most – LMR-LRT (p) .010 .002 .005 .229 classes, but the female prolonged grief and moderate-stable classes showed B-LRT (p) – <.001 <.001 <.001 <.001 a percentual increase across time. AIC, Akaike Information Criterion; BIC, Bayesian Information Criterion; B-LRT, bootstrapped likelihood ratio test; LMR-LRT, Lo-Mendell-Rubin likelihood ratio 3.4.4. Predictors of class membership test; SSA-BIC, Sample Size Adjusted BIC. Covariates (i.e., optimism, neuroticism, mental health, and physical Best-fitting model indicated in bold font. health) were nested within the four-class model to investigate predictors of class membership. The first set of analyses used the covariates as ¼ significantlyworse model fitfor the five-classmodel solution (p .229). predictors and the four classes as outcomes (Table 4). Compared to the Given these indices, the optimal solution was the four-class model with resilient class, the moderate-stable class showed lower mental and phys­ random intercept and fixed slope. ical health, the prolonged grief class showed lower mental health and In the four-class model (see Fig. 1), the majority of the sample fol­ optimism, while the recovery class showed lower mental health. lowed a resilient (64.4%) trajectory characterized by low-level PGD Compared to the prolonged grief class, the moderate-stable and the re­ ¼ � < symptoms at T1 (intercept M 2.16 0.13, p .001) with a slight covery classes showed lower mental health. Comparisons between ¼ À � < decline over time (slope M 0.14 0.01, p .001). The second- moderate-stable and recovery classes did not reach statistical significance. largest class, moderate-stable (20.4%), showed a moderate symptom Analyses with interactions between gender and psychosocial covariates ¼ � < level at T1 (intercept M 4.50 0.27, p .001) and an almost flat included did not reach statistical significance; that is, associations of ¼ À � ¼ growth curve (slope M 0.01 0.03, p .644). A third class, recovery class membership and covariates did not differ according to gender. (8.4%), endorsed high baseline grief symptoms (intercept M ¼ 6.58 � < ¼ À � 0.54, p .001) and a steady time-wise decline (slope M 0.43 0.04, 4. Discussion p < .001). The fourth class was characterized by high-level prolonged ¼ � < grief (6.8%) symptoms at T1 (intercept M 7.77 0.42, p .001), Given the adverse effects associated with spousal bereavement ¼ À � < showing minimal decrease over time (slope M 0.04 0.05, p (Stroebe et al., 2007), knowledge of variations in risk towards .487). loss-related complications is critical. Gender might account for some of the variability, but studies on gender differences within the newly 3.4. Gender differences introduced ICD-11 diagnosis of PGD is lacking. With symptom duration as a critical distinguisher between normal and pathological grief re­ 3.4.1. Multiple-group GMM actions, GMM is a particularly suitable method to explore possible Within the four-class model, gender was used as a known-class var­ gender variations within the bereavement field.The present study is the iable to evaluate gender-specific differences in trajectory parameters. first to explore gender differences in ICD-11 PGD within a GMM The multiple-group GMM analysis demonstrated high entropy (0.88). framework. The overall Wald Test revealed significant differences in trajectory pa­ Our analyses revealed four trajectories with different symptomatic < rameters for men and women (p .05), suggesting that the known-class patterns, indicating considerable variation in the natural course of grief variable increased model fit (see Supplementary Table 2). Focused symptoms. While the resilient, moderate-stable, and recovery classes did comparisons indicated significantbetween-gender variation in intercept not differ according to gender, men and women in the prolonged grief ¼ ¼ (p .018) and slope (p .004) of the prolonged grief class. No other trajectory demonstrated different symptom profiles. In this class, men significant parameter differences emerged. Fig. 2 illustrates almost expressed significantlymore baseline grief symptomatology (i.e., higher identical growth patterns for male and female resilient, moderate-stable, intercept) compared to women. Symptom development also differed in and recovery classes, while the prolonged grief classes visibly differ be­ 1 the prolonged grief class as men expressed symptom decrease (i.e., tween genders. Exploratory analyses examined whether specific negative slope), while women expressed symptom increase (i.e., positive symptoms contributed to this difference. Men in the prolonged grief class slope). The nature of this difference is uncertain. Exploratory analyses “ ’ ” ¼ endorsed symptoms feeling one has lost a part of one s self (p .009) indicated that men in the prolonged grief class more often than women “ ” ¼ and engagement difficulties (p .004) significantly more often than endorsed the PGD symptom “difficultyin engaging with social or other “ ” women at T1 and symptoms inability to experience positive mood (p activities” at two and six months post-loss, while no gender difference ¼ “ ” ¼ .008) and engagement difficulties (p .003) more often at T2. appeared at eleven months. Hypothetically, men in the prolonged grief class experience acute engagement difficulties, but over time – as 3.4.2. Proportions across trajectories engagement improves or does not cause difficulties – symptom Comparisons of gender prevalence rates within the different classes endorsement resembles the female level. Perhaps, this change may > were non-significant (ps .05); i.e., men and women did not differ in contribute to explain the overall symptom decline (i.e., negative slope) membership proportions across trajectories. in the male class. Considering this hypothesis, varied symptom endorsement at T3 was expectable, reflecting the female prolonged grief symptom increase (i.e., positive slope). Yet, no symptoms were endorsed significantlymore often by women. Given the highly exploratory nature of these analyses, further research could investigate how symptom 1 Four-class GMM analyses on the male and female subsamples separately resulted in comparable trajectory patterns as in the multiple-group model endorsement contributes to gender differences in symptomatic devel­ [male- and female-only four-class figures available upon request]. Multiple- opment of PGD. group GMM was conducted to enable focused comparisons of gender-specific Similar proportions of men and women received membership of the trajectory parameters (i.e., slope and intercept) within the same model, respective trajectories. Consistent with previous research (Bonanno expanding knowledge beyond separated models. et al., 2005; Bonanno and Malgaroli, 2020; Galatzer-Levy and Bonanno,

171 M. Lundorff et al. Journal of Psychiatric Research 129 (2020) 168–175

Fig. 1. Four-class unconditional trajectory model of ICD-11 prolonged grief disorder symptoms (N ¼ 857).

2012; Maccallum et al., 2015), the majority of both men (68.0%) and Thus, whereas this class develops differently across time for men and women (65.1%) followed a resilient trajectory characterized by a con­ women, the presence of elevated PGD symptoms are similarly frequent. stant low symptom-level. Although spousal loss is a significantlife event This finding appears in agreement with earlier studies, demonstrating (Holmes and Rahe, 1967), most individuals are seemingly capable of how prevalence rates of disturbed grief do not differ according to gender coping effectively with the associated distress. Comparable proportions (Djelantik et al., 2020; Lundorff et al., 2017). of men (5.8%) and women (5.2%) followed the prolonged grief trajectory. The identified trajectories differed in their ability to capture PGD

Fig. 2. Multiple-group (gender) four-class trajectory model of ICD-11 prolonged grief disorder symptoms (N ¼ 857).

172 M. Lundorff et al. Journal of Psychiatric Research 129 (2020) 168–175

Table 3 post-loss (T2) and in the female prolonged grief class at eleven months Diagnostic PGD cases within prolonged grief symptomatic trajectories. (T3). The multiple-group trajectory model illustrates this visually as Males Females male and female prolonged grief demonstrate symptom peak at different times. From a diagnostic perspective, men were most likely to exhibit PGD at T2, PGD at T3, PGD at T2, % PGD at T3, % % (n) % (n) (n) (n) complications resembling probable PGD around six months post-loss, subsequently declining. At the same time, a small female minority Resilient 6.8 (11) 1.3 (2) 6.8 (24) 3.2 (11) Moderate-stable 30.6 (15) 29.2 (14) 37.4 (43) 37.5 (42) experienced worsening symptom-severity and increasing probable case Recovery 27.3 (3) 0 (0) 44.2 (19) 14.0 (6) rate across time. Considering time-frames used for diagnosing grief, this Prolonged grief 84.6 (11) 45.5 (5) 67.7 (21) 74.1 (20) might explain why past research identifiesfemale gender as a risk factor Fisher’s Exact 49.39 (234) 44.02 (228) 112.79 (540) 132.62 (523) for developing pathological grief reactions (Burke and Neimeyer, 2013; Test (N) Heeke et al., 2017). Given the current study’s reliance on self-report p <.001 <.001 <.001 <.001 data, additional research should explore the complex relationship be­ PGD, prolonged grief disorder; T2, time 2 (six months post-loss); T3, time 3 tween gender, timing, and diagnosing PGD. (eleven months post-loss). Low optimism measured at two months following bereavement PGD cases reported the presence of at least one core symptom (longing or pre­ predicted membership of the PGD-frequent trajectory (viz., prolonged occupation), at least one symptom of emotional pain, and significantfunctional grief). Optimism reflects favorable expectations for the future (Carver impairment. Adjusted sample sizes reported due to missing values on the diagnostic test for et al., 2010). Corresponding to earlier findings (e.g., Golden and Dal­ probable ICD-11 PGD. gleish, 2012; Thomsen et al., 2018), the absence of positivity in future Cells with �5% observations indicated in bold font. thinking might be a key marker of later bereavement-related distress. Neuroticism – reflecting a tendency towards intense and frequent negative emotions and thoughts (McCrae and Costa, 2008) – has pre­ Table 4 viously been identified as a predictor of pathological grief (Burke and Multinomial logistic regression estimates for predictors of class membership. Neimeyer, 2013). Contrary to expectations, neuroticism did not predict Covariate Est SE p membership in the prolonged grief class. Although considered an enduring indicator of emotional stability (Costa and McCrae, 2011), our Compared to resilient Moderate-stable Physical health À 0.041 0.016 .008 study assessed neuroticism fairly early following bereavement, at which Mental health À 0.140 0.031 .000 point statements such as “I am rarely depressed or sad” might be Optimism À 0.064 0.053 .226 endorsed more frequently. As negative emotions are a natural part of À Neuroticism 0.006 0.028 .821 acute grief for many, even healthy individuals (Zisook and Shear, 2009), Prolonged grief Physical health À 0.090 0.047 .052 Mental health À 0.323 0.060 .000 the insignificant association between neuroticism and PGD trajectories Optimism À 0.326 0.162 .044 might reflect a generally deflated mood among the participants. Low Neuroticism 0.010 0.060 .873 physical health significantly predicted the moderate-stable class, which Recovery Physical health À 0.040 0.022 .076 also included substantial proportions of probable PGD cases, and Mental health À 0.178 0.028 .000 approached significance as a predictor of the prolonged grief class (p ¼ Optimism À 0.098 0.063 .120 Neuroticism À 0.007 0.032 .831 .052). This finding suggests how severe PGD symptoms may entail Compared to prolonged grief physical constraints. The resilient class reported higher mental health Recovery Physical health 0.051 0.045 .258 compared to all other classes, indicating how healthy mental states Mental health 0.145 0.065 .026 enable better coping with grief. Relatedly, studies demonstrate how Optimism 0.229 0.161 .155 Neuroticism À 0.016 0.053 .759 pathological grief predicts future lower mental health (Boelen and Pri­ Moderate-stable Physical health 0.049 0.044 .262 gerson, 2007) and lower functioning (Bonanno et al., 2007). This Mental health 0.182 0.046 .000 tentatively indicates a bidirectional relationship; restricted mental Optimism 0.263 0.153 .085 health incapacitates resilient grieving, accelerating PGD-related Neuroticism À 0.016 0.063 .803 distress, leading to a further decrease in mental well-being. Notably, Compared to moderate-stable Recovery Physical health 0.001 0.023 .952 interactions between these predictors and gender were non-significant, Mental health À 0.038 0.041 .361 indicating that the tested covariates did not systematically differ in Optimism À 0.034 0.062 .584 their ability to predict trajectories across men and women. À Neuroticism 0.001 0.030 .984 The study has numerous strengths, including its registry-based Est, Parameter Estimate; SE, Standard Error. recruitment. The trajectory analysis relied on the official ICD-11 PGD Statistically significant values indicated in bold font. criteria, and the large study cohort enabled subgroup exploration. Importantly, the study’s primary outcome (i.e., PGD symptom-severity) cases. Few individuals in the resilient class, regardless of gender, met did not significantly differ between the male and female subsamples, ICD-11 PGD criteria, resembling findingsfrom earlier research (Bonanno making the sample suitable for studying subgroup-specific gender dif­ and Malgaroli, 2020). Six months post-loss, the prolonged grief class ferences. Hence, the study is the first to explore gender differences in best-captured caseness as 84.6% of men and 67.7% of women demon­ ICD-11 PGD symptom development across early bereavement within a strated criteria for probable ICD-11 PGD. However, the recovery and the GMM framework. Still, some limitations should be mentioned. First, moderate-stable classes also captured substantial proportions of PGD predictors were measured two months into bereavement, at which point cases in both genders six months into bereavement. Eleven months many individuals are still affected by their recent loss, potentially post-loss, the prolonged grief class, characterized by elevated symptoms, influencing the stability of otherwise fixed constructs. Additional similarly best captured PGD caseness. This findingaligns with the notion research, preferably with pre-morbid measures, is needed to expand that the central distinguishing feature of PGD is the persistent, intense knowledge on which psychosocial factors distinguish normative from grief symptoms (Maciejewski et al., 2016; Maercker et al., 2013). Within more pathological classes of ICD-11 PGD. Second, although the total the prolonged grief trajectory, an important gender difference emerged; sample size was sufficient for GMM analyses, the limited power of the eleven months post-loss PGD proportions had dropped to 45.5% among smaller gender-subgroups (especially in the prolonged grief trajectory) men and increased to 74.1% in the female subsample. Accordingly, PGD might explain the non-significant interactions between gender and cases were most frequent in the male prolonged grief class at six months covariates. Results from the multiple-group GMM should be interpreted with caution. Third, PGD assessment relied on items combined from

173 M. Lundorff et al. Journal of Psychiatric Research 129 (2020) 168–175 different self-report measures. Without structured clinical interviews, Bonanno, G.A., 2004. Loss, trauma, and human resilience: have we underestimated the only probable PGD caseness is assigned. In continuation of this, the human capacity to thrive after extremely aversive events? Am. Psychol. https://doi. org/10.1037/0003-066X.59.1.20. GMM analysis did not distinguish between core and accompanying Bonanno, G.A., Kaltman, S., 2001. The varieties of grief experience. Clin. Psychol. Rev. emotional pain symptoms of PGD. Still, we found high internal consis­ 21, 705–734. https://doi.org/10.1016/S0272-7358(00)00062-3. tency for our PGD measure at T1-T3 (α ¼ 0.89-0.90). Bonanno, G.A., Malgaroli, M., 2020. Trajectories of grief: comparing symptoms from the DSM-5 and ICD-11 diagnoses. Depress. Anxiety 37, 17–25. https://doi.org/10.1002/ Notwithstanding these limitations, the present study revealed gender da.22902. similarities and differences in ICD-11 PGD symptomatology during the Bonanno, G.A., Neria, Y., Mancini, A., Coifman, K.G., Litz, B., Insel, B., 2007. Is there firstyear of spousal bereavement. The vast majority of men and women more to complicated grief than depression and posttraumatic stress disorder? A test of incremental validity. J. Abnorm. Psychol. 116, 342–351. https://doi.org/ followed a resilient trajectory with a continuous low grief-level, con­ 10.1037/0021-843X.116.2.342. firming this is the modal response to spousal bereavement. While a Bonanno, G.A., Papa, A., Moskowitz, J.T., Folkman, S., 2005. Resilience to loss in comparable minority of both genders followed a prolonged grief trajec­ bereaved spouses, bereaved parents, and bereaved gay men. J. Pers. Soc. Psychol. ’ https://doi.org/10.1037/0022-3514.88.5.827. tory characterized by high-level PGD symptoms, this class symptomatic Bonanno, G.A., Wortman, C.B., Lehman, D.R., Tweed, R.G., Haring, M., Sonnega, J., pattern differed between men and women. Men expressed an acute, Carr, D., Nesse, R.M., 2002. Resilience to loss and chronic grief: a prospective study decreasing grief reaction, whereas women showed an adjourned, from preloss to 18-months postloss. J. Pers. Soc. Psychol. 83, 1150–1164. https:// increasing response. Applying a diagnostic algorithm for PGD, the pro­ doi.org/10.1037/0022-3514.83.5.1150. Burke, L.A., Neimeyer, R.A., 2013. Prospective risk factors for complicated grief: a longed grief trajectory best captured diagnostic caseness and was pre­ review of the empirical literature. In: Stroebe, M.S., Schut, H.A.W., Van den Bout, J. dicted by low optimism and low mental health. From a clinical (Eds.), Complicated Grief. Scientific Foundations for Health Care Professionals. – perspective, findings indicate how men and women show time- Routledge, New York, NY, pp. 145 161. Carver, C.S., Scheier, M.F., Segerstrom, S.C., 2010. Optimism. Clin. Psychol. Rev. https:// dependent differences in their vulnerability towards PGD symptoms. doi.org/10.1016/j.cpr.2010.01.006. The study provides valuable knowledge before the ICD-11 PGD imple­ Clark, S., Muth�en, B., 2009. Relating Latent Class Analysis Results to Variables Not mentation by suggesting how gender may account for decreasing and Included in the Analysis [WWW Document]. Univ. Calif. Los Angeles. https://doi. org/10.1016/j.snb.2016.06.061. increasing symptom development in early spousal bereavement. Costa, P.T., McCrae, R.R., 2011. Revised NEO personality inventory (NEO-PI-R) and NEO five-factor inventory (NEO-FFI) professional manual. In: Archer, R.P., Smith, S.R. (Eds.). Routledge/Taylor & Francis Group, New York, NY, pp. 223–256. Contributions Costa, P.T., McCrae, R.R., 1992. Revised NEO personality inventory (NEO-PI-R) and NEO five-factor inventory (NEO-FFI) professional manual. Psychol. Assess. Resour. M. Lundorff, G.A. Bonanno, M. Johannsen, and M. O’Connor Djelantik, A.A.A.M.J., Smid, G.E., Mroz, A., Kleber, R.J., Boelen, P.A., 2020. The prevalence of prolonged grief disorder in bereaved individuals following unnatural designed the study. M. Lundorff and G.A. Bonanno analyzed and inter­ losses: systematic review and meta regression analysis. J. Affect. Disord. https://doi. preted the data. M. Lundorff wrote the manuscript. All authors have org/10.1016/j.jad.2020.01.034. approved the final manuscript. Galatzer-Levy, I.R., Bonanno, G.A., 2012. Beyond normality in the study of bereavement: heterogeneity in depression outcomes following loss in older adults. Soc. Sci. Med. https://doi.org/10.1016/j.socscimed.2012.02.022. Declaration of conflicting interest (COI) Golden, A.M.J., Dalgleish, T., 2012. Facets of pejorative self-processing in complicated grief. J. Consult. Clin. Psychol. https://doi.org/10.1037/a0027338. Harris, P.A., Taylor, R., Thielke, R., Payne, J., Gonzalez, N., Conde, J.G., 2009. Research None. electronic data capture (REDCap)-A metadata-driven methodology and workflow process for providing translational research informatics support. J. Biomed. Inf. 42, 377–381. https://doi.org/10.1016/j.jbi.2008.08.010. Role of funding source Heeke, C., Kampisiou, C., Niemeyer, H., Knaevelsrud, C., 2017. A systematic review and meta-analysis of correlates of prolonged grief disorder in adults exposed to violent The Aarhus Bereavement Study was funded by a grant from the loss. Eur. J. Psychotraumatol. https://doi.org/10.1080/20008198.2019.1583524. À Holmes, T.H., Rahe, R.H., 1967. The social readjustment rating scale. J. Psychosom. Res. Aarhus University Research Foundation, Denmark (AUFF-E 2015-FLS- 11, 213–218. https://doi.org/10.1016/0022-3999(67)90010-4. 8-63) awarded to the last author. The funder had no role in study design, IBM Corporation, 2019. IBM SPSS Statistics for Windows. data collection and analysis, decision to publish, or preparation of the Killikelly, C., Lorenz, L., Bauer, S., Mahat-Shamir, M., Ben-Ezra, M., Maercker, A., 2019. Prolonged grief disorder: its co-occurrence with adjustment disorder and post- manuscript. traumatic stress disorder in a bereaved Israeli general-population sample. J. Affect. Disord. https://doi.org/10.1016/j.jad.2019.02.014. Acknowledgements Lenferink, L.I.M., Nickerson, A., de Keijser, J., Smid, G.E., Boelen, P.A., 2018. Trajectories of grief, depression, and posttraumatic stress in disaster-bereaved people. Depress. Anxiety. https://doi.org/10.1002/da.22850. We are indebted to The Aarhus Bereavement Study (TABstudy) Lundorff, M., Holmgren, H., Zachariae, R., Farver-Vestergaard, I., O’Connor, M., 2017. participants, without whom these analyses would not be possible. Prevalence of prolonged grief disorder in adult bereavement: a systematic review and meta-analysis. J. Affect. Disord. 212, 138–149. https://doi.org/10.1016/j. jad.2017.01.030. Appendix A. Supplementary data Maccallum, F., Galatzer-Levy, I.R., Bonanno, G.A., 2015. Trajectories of depression following spousal and child bereavement: a comparison of the heterogeneity in outcomes. J. Psychiatr. Res. https://doi.org/10.1016/j.jpsychires.2015.07.017. Supplementary data to this article can be found online at https://doi. Maciejewski, P.K., Maercker, A., Boelen, P.A., Prigerson, H.G., 2016. “Prolonged grief org/10.1016/j.jpsychires.2020.06.030. disorder” and “persistent complex bereavement disorder”, but not “complicated grief”, are one and the same diagnostic entity: an analysis of data from the Yale Bereavement Study. World Psychiatr. 15, 266–275. https://doi.org/10.1002/ References wps.20348. Maercker, A., Brewin, C.R., Bryant, R.A., Cloitre, M., van Ommeren, M., Jones, L.M., Humayan, A., Kagee, A., Llosa, A.E., Rousseau, C., Somasundaram, D.J., Souza, R., American Psychiatric Association, 2013. In: Arlington, V.A. (Ed.), Diagnostic and Suzuki, Y., Weissbecker, I., Wessely, S.C., First, M.B., Reed, G.M., 2013. Diagnosis Statistical Manual of Mental Disorders, p. 5. https://doi.org/10.1176/appi. and classification of disorders specifically associated with stress: proposals for ICD- books.9780890425596. 11. World Psychiatr. 12, 198–206. https://doi.org/10.1002/wps.20057. Boelen, P.A., Lenferink, L.I.M., Nickerson, A., Smid, G.E., 2018. Evaluation of the factor McCrae, R.R., Costa, P.T.J., 2008. The five-factor theory of personality. In: Handbook of structure, prevalence, and validity of disturbed grief in DSM-5 and ICD-11. J. Affect. Personality: Theory and Research. Disord. 240, 79–87. https://doi.org/10.1016/j.jad.2018.07.041. Muthen,� B., 2003. Statistical and substantive checking in growth mixture modeling: Boelen, P.A., Lenferink, L.I.M., Smid, G.E., 2019. Further evaluation of the factor comment on bauer and curran (2003). Psychol. Methods. https://doi.org/10.1037/ structure, prevalence, and concurrent validity of DSM-5 criteria for Persistent 1082-989X.8.3.369. Complex Bereavement Disorder and ICD-11 criteria for Prolonged Grief Disorder. Muthen,� B., Muth�en, L., 2018. Mplus. Psychiatr. Res. 273, 206–210. https://doi.org/10.1016/j.psychres.2019.01.006. Nagin, D., Odgers, C.L., 2010. Group-based trajectory modeling in clinical research. Boelen, P.A., Prigerson, H.G., 2007. The influence of symptoms of prolonged grief SSRN. https://doi.org/10.1146/annurev.clinpsy.121208.131413. disorder, depression, and anxiety on quality of life among bereaved adults. Eur. Arch. Psychiatr. Clin. Neurosci. 257, 444–452. https://doi.org/10.1007/s00406- 007-0744-0.

174 M. Lundorff et al. Journal of Psychiatric Research 129 (2020) 168–175

Nielsen, M.K., Carlsen, A.H., Neergaard, M.A., Bidstrup, P.E., Guldin, M.-B., 2018. Smith, K.V., Ehlers, A., 2019. Cognitive predictors of grief trajectories in the firstmonths Looking beyond the mean in grief trajectories: a prospective, population-based of loss: a latent growth mixture model. J. Consult. Clin. Psychol. undefined, cohort study. Soc. Sci. Med. https://doi.org/10.1016/J.SOCSCIMED.2018.10.007. undefined. https://doi.org/10.1037/ccp0000438. Prigerson, H.G., Horowitz, M.J., Jacobs, S.C., Goodkin, C., Parkes, C.M., Aslan, M., Stroebe, M.S., Schut, H.A.W., Stroebe, W., 2007. Health outcomes of bereavement. Raphael, B., Marwit, S.J., Wortman, C., Neimeyer, R.A., Bonanno, G.A., Block, S.D., Lancet 370, 1960–1973. https://doi.org/10.1016/S0140-6736(07)61816-9. Kissane, D., Boelen, P.A., Maercker, A., Litz, B.T., Johnson, J.G., First, M.B., Sullivan, G.M., Feinn, R., 2012. Using effect size—or why the P value is not enough. Maciejewski, P.K., 2009. Prolonged grief disorder: psychometric validation of J. Grad. Med. Educ. 4, 279–282. https://doi.org/10.4300/JGME-D-12-00156.1. criteria proposed for DSM-V and ICD-11. PLoS Med. 6, 1–12. https://doi.org/ Sveen, J., Bergh Johannesson, K., Cernvall, M., Arnberg, F.K., 2018. Trajectories of 10.1371/journal.pmed.1000121. prolonged grief one to six years after a natural disaster. PLoS One 13, e0209757. Prigerson, H.G., Jacobs, S.C., 2001. Traumatic grief as a distinct disorder: a rationale, https://doi.org/10.1371/journal.pone.0209757. consensus criteria, and a preliminary empirical test. In: Stroebe, M.S., Hansson, R.O. Thomsen, D.K., Lundorff, M., Damkier, A., O’Connor, M., 2018. Narrative identity and (Eds.), Handbook of Bereavement Research: Consequences, Coping, and Care. grief reactions: a prospective study of bereaved partners. J. Appl. Res. Mem. Cogn. American Psychological Association, Washington, D.C., D.C., pp. 614–646 https://doi.org/10.1016/J.JARMAC.2018.03.011. Ruggiero, K.J., Ben, K. Del, Scotti, J.R., Rabalais, A.E., 2003. Psychometric properties of Ware, J.E., Kosinski, M., Keller, S.D., 1996. A 12-Item Short-Form Health Survey: the PTSD checklist—civilian version. J. Trauma Stress 16, 495–502. https://doi.org/ construction of scales and preliminary tests of reliability and validity. Med. Care 34, 10.1023/A:1025714729117. 220–233. https://doi.org/10.1097/00005650-199603000-00003. Scheier, M.F., Carver, C.S., Bridges, M.W., 1994. Distinguishing optimism from World Health Organization, 2018. ICD-11 for mortality and morbidity statistics (version neuroticism (and trait anxiety, self-mastery, and self-esteem): a reevaluation of the for preparing implementation) [WWW document]. 6B42 prolong. Grief disord. URL. life orientation test. J. Pers. Soc. Psychol. 67, 1063–1078. https://doi.org/10.1037/ https://icd.who.int/browse11/l-m/en#/http://id.who.int/icd/entity/1183832314. 0022-3514.67.6.1063. Yu, C.-Y, 2002. Evaluating Cutoff Criteria of Model Fit Indices for Latent Variable Models Shear, M.K., Simon, N., Wall, M., Zisook, S., Neimeyer, R., Duan, N., Reynolds, C., with Binary and Continuous Outcomes. Unpublished doctoral dissertation. Lebowitz, B., Sung, S., Ghesquiere, A., Gorscak, B., Clayton, P., Ito, M., Nakajima, S., University of California, Los Angeles. Konishi, T., Melhem, N., Meert, K., Schiff, M., O’Connor, M.-F., First, M.B., Zhou, N., Wen, J., Stelzer, E.M., Killikelly, C., Yu, W., Xu, X., Shi, G., Luo, H., Wang, J., Sareen, J., Bolton, J., Skritskaya, N., Mancini, A.D., Keshaviah, A., 2011. Maercker, A., 2020. Prevalence and associated factors of prolonged grief disorder in Complicated grief and related bereavement issues for DSM-5. Depress. Anxiety 28, Chinese parents bereaved by losing their only child. Psychiatr. Res. https://doi.org/ 103–117. https://doi.org/10.1002/da.20780. 10.1016/j.psychres.2020.112766. Zisook, S., Shear, M.K., 2009. Grief and bereavement: what psychiatrists need to know. World Psychiatr. 8, 67–74.

175